Choosing between AI video generators and code-based tools for your video effects? Here’s the quick way to decide before you burn hours testing everything yourself. I’ve been staring at so many “AI video” tutorials lately, and most of them push you toward making fully AI-generated clips that feel hollow. So I was thrilled when I found a breakdown that does the opposite.
This whole walkthrough comes from Matt Wolfe, the creator behind Future Tools and those wild intros on his Friday news videos. He’s not selling a fully-automated future. His whole pitch is 95% human video, 5% AI spice, and I think that framing is exactly right for anyone who still wants their content to feel real. He shares the exact tools, prompts, and steps he uses, so let’s break down how to choose the right one for each job.
The core decision: generators vs code tools
The author splits his toolkit into two camps, and knowing which camp fits your task saves you the most time.
- 🎬 AI video generators (Runway, Seed Dance 2.0, Kling, Veo 3.1, Google’s Gemini/Omni): great for cinematic, physics-heavy effects like bursting through a wall, wormhole transitions, or a Yeti walking behind you.
- 🧩 Code-based tools (Remotion inside Codex or Claude Code, plus Claude Co-work/Opus): great for anything needing perfect text, accurate data, highlights, and logo reveals.
Here’s the mental model he lands on: if the effect needs to look epic and a little unpredictable, reach for a generator. If it needs to be exact, reach for code.
Pros and cons, straight from his testing
AI generators
- Pros: stunning motion, easy keyframe workflow (upload a first frame and a last frame, add a prompt), and the models are trained on tons of stock-style footage so B-roll looks convincing.
- Cons: text comes out garbled (he got “CEO of Busty Bench” instead of his real title), and geography is a mess. His “plane from San Diego to New York” test flew over made-up states and overshot the city.
Code tools (Remotion / Claude Co-work)
- Pros: nails text every single time, produces accurate charts and maps, and gives you a clean green screen background for compositing.
- Cons: less flashy animation. The mind behind this admits Remotion “nailed the text” but wasn’t as cinematic.
His recommendation
The expert’s rule of thumb is simple. Use generators for spectacle, code for precision. When he needed a lower third with his name spelled right, Veo 3.1 beat Seed Dance, but Remotion beat them both for guaranteed accuracy. When he wanted a particle explosion reforming into his logo, Remotion (with a skill called “remotion best practices”) was the pick. And for effects that happen around you without you reacting, Google’s Omni model won.
I love that he doesn’t crown one winner. He matches the tool to the outcome, which is the whole point.
The signature intro, step by step
This is the trick people ask him about most, and the process is refreshingly doable:
- Record yourself starting off-camera. Hit record while reaching in, so you capture an empty-room frame before you sit down.
- In DaVinci Resolve, export two stills: “frame one” (empty room) and “frame two” (right before you speak).
- In Runway, pick Custom, then Keyframe. Upload frame one as the first frame and frame two as the last.
- Prompt it simply, like “the man bursts through the back wall and then sits down in the chair.”
- Use Seed Dance 2.0 (his favorite), output with audio on so you can mute later, and generate.
- Back in DaVinci, splice it in and use the “smooth cut” transition to blend the seam.
He notes you’ll often re-roll a few times to kill hard cuts, and that these same models live inside Krea, Higgsfield, Leonardo, and more. He just happens to have Runway credits.
Practical tips worth stealing
- Plan transitions in advance. For a location swap to a Big Bear cabin, he filmed the intro at home first, wore the same shirt in both spots, and prompted the camera to rotate from one scene to the other.
- Add “the man does not speak” to transition prompts, or the model makes you mumble gibberish.
- For background effects, feed a 10-second clip into Gemini and ask for a Yeti, an explosion, or even a weather change from sunny to rainy. Add “make it so I don’t react” so you don’t turn around.
- For text B-roll, Claude Co-work (Opus 4.8) or Fable can screenshot a webpage, scroll to a paragraph, and add a highlighter animation, all by writing code behind the scenes. Fable came out a touch smoother.
- For the talking-head trick (his Sam Altman and Mira Murati text exchange), he read both sides out loud, recorded the audio, and used Runway’s “character script to video” with a still image plus audio upload. Then he cut the two clips together as a conversation.
The honest caveats
This contributor is upfront about the rough edges. Animal-to-human morphs still look bad. City-to-city maps are unreliable across every generator he tried. And text in generators is a coin flip. His fix is either re-prompting four or five times or switching to a code tool when accuracy matters. For infographics and explainers, he even pulls animations from Notebook LM’s video overview feature, which builds surprisingly usable B-roll from your source docs.
The bigger point he makes stuck with me: he purposely shows off that something is AI, so viewers never wonder. In a feed where telling real from fake gets harder daily, that honesty is a feature, not a limitation.
Want the exact prompts, the claw-machine and wolf intro variations, and the on-screen demos? Go watch the full video. It’s the clearest tool-by-tool map I’ve seen for adding real spice without turning your channel into slop.